How “consciousness” allows a cognitive tutoring agent make good diagnosis during astronauts' training

Striving in the real world is more and more what artificial agents are required to do, and it is not a simple task. Interacting with humans in general, and with students in specific, requires an awful lot of subtlety if one is to be perceived as a great tutor and a pleasant fellow. Similarly, the more various types of information an artificial agent senses, the more apt it may be. But then comes the need to process all this stuff, and that can overwhelm even the most powerful computer. «Consciousness» mechanisms can help and sustain an apt tutor, allowing it to consider various sources of information in diagnosing and guiding learners. We show in the present paper how they effectively support theses processes in the specific context of astronauts training on the manipulation of the Space Station Robotic Manipulation System, Canadarm2.